Analyzing Handwriting Biometrics in Metadata Context

Tobias Scheidat, Franziska Wolf, Claus Vielhauer
Dept. of Computer Science, Univ. of Magdeburg, Universitätsplatz 2, 39106 Magdeburg, Germany


ABSTRACT

In this article, methods for user recognition by online handwriting are experimentally analyzed using a combination of
demographic data of users in relation to their handwriting habits. Online handwriting as a biometric method is
characterized by having high variations of characteristics that influences the reliance and security of this method. These
variations have not been researched in detail so far. Especially in cross-cultural application it is urgent to reveal the
impact of personal background to security aspects in biometrics. Metadata represent the background of writers, by
introducing cultural, biological and conditional (changing) aspects like fist language, country of origin, gender,
handedness, experiences the influence handwriting and language skills. The goal is the revelation of intercultural impacts
on handwriting in order to achieve higher security in biometrical systems. In our experiments, in order to achieve a
relatively high coverage, 48 different handwriting tasks have been accomplished by 47 users from three countries
(Germany, India and Italy) have been investigated with respect to the relations of metadata and biometric recognition
performance. For this purpose, hypotheses have been formulated and have been evaluated using the measurement of
well-known recognition error rates from biometrics. The evaluation addressed both: system reliance and security threads
by skilled forgeries. For the later purpose, a novel forgery type is introduced, which applies the personal metadata to
security aspects and includes new methods of security tests. Finally in our paper, we formulate recommendations for
specific user groups and handwriting samples.

Keywords: biometrics, cross-cultural, metadata, online handwriting, skilled attacks, soft biometrics, verification

1. INTRODUCTION

Active methods in biometrics like voice and handwriting have achieved a new powerful position in user authentication
systems recently. Previous researches already introduced to new methods of analyzing active biometric methods (voice
and handwriting) in context of metadata. For example in handwriting recognition in [1] text is analyzed by using global
image features and image indexing by automatic page analysis and segmentation. Image based on image-to-image
similarity measure and text based on text-to-image score are used for retrieval of data. Metadata that describes both,
technical and personal characteristics in order to relate both facts to the context of speech-based user authentication was
introduced in [2]. Methods of analyzing handwritten documents regarding aspects of person related data like gender or
ethic background have been made. In [3] handwriting of Indian users was analyzed for identification performance and
quality based on demographic information about the writer. Different characters were ranked and the individual
performance of characters for group identification was measured. Based on demographic data of gender an age the
accumulated performance of characters could be achieved between 65% and 85%. Nevertheless only static type faces of
handwriting have been researched so far. In our research as shown in recent work [4] we refer to dynamic handwriting
data and relate this to metadata in order to investigate cultural impacts to enhance authentication security. Analysis on
online handwriting data metadata can be used to enhance the evaluation of handwriting in intercultural context.
We now also introduce a new area of security application with respect to forgery levels. Metadata include contextual
information regarding the experimental environment, as well as for the linguistic, cultural, ethic and educational
background of a person, which is attributed to the biometric data. These new aspects of evaluating voice and handwriting
considering cultural impacts like language, script and nationality are of special interest to technology. For example,
different cultural background may affect usability and security (i.e. recognition accuracy) of a biometric system in
international cross-cultural and cross-lingual context. In our previous work, we have presented a novel framework
consisting of a metadata model and an acquisition methodology for an experimental environment based on the biometric
modalities of speech and voice [5]. For the domain of face recognition, Phillips et al. describe in [6] a procedure and
results in the Face Recognition Vendor Test 2002. For the tests a very large data set of 37,437 individuals was used and
the data were examined also from so-called demographic aspects. For the best systems the identification rates for males
were 6% to 9% points higher than that of females. Another result of the tests was that identification of older people is
Security, Steganography, and Watermarking of Multimedia Contents VIII, edited by Edward J. Delp III, Ping Wah Wong,
Proc. of SPIE-IS&T Electronic Imaging, SPIE Vol. 6072, 60720H, © 2006 SPIE-IS&T · 0277-786X/06/$15
SPIE-IS&T/ Vol. 6072 60720H-1

easier than of younger people. The identification performance increases approximately 5% points for every ten years
increase of age. These outcomes motivate further examinations in the area of metadata (here gender and age) also for
other biometric traits like online handwriting.
New aspects of metadata based security analysis with respect to cultural and biological and new conditional metadata
can be formulated. The main interest in both areas of research for us is the definition of user groups based on their
metadata (e.g. by gender or native language). These groups are analyzed and compared by aspects of handwriting
classified into relation of contend (here called semantic) and relation of data parameters (here called syntax) . The goal is
to formulate metadata based recommendations for usability and security, which may enhance active biometric methods
in future. The intercultural data collection is gained from the research project CultureTech ([7]), having project partners
in Germany, Italy and India. Standardized handwriting samples and metadata have been collected in order to have a
decent base for an analysis with focus on intercultural matters.
By analyzing both, user group’s metadata spreading and their handwriting parameters, hypotheses are formulated that
relate the metadata by intercultural aspects to the handwriting specifics. Evaluations can be done by referring to a well-
established recognition measurement in biometrics, equal error rates (EER), with respect to inner and outer safety. Here,
inner safety refers to verification and random attacks declares the system’s security without the thread of forgery,
whereas outer safety refers to varying attacks based on a hierarchy of sophistication. Even though the subject quantity in
our evaluations cannot claim statistical completeness, bias estimations can be gained using these results and differing
commendations concerning using samples can be formulated in order to achieve optimal security levels for each subject
group due to metadata.
Figure 1 shows the general process of authentication based on enrollment data and test data. First, the enrollment data,
consisting of handwriting data and metadata, is collected, in order to register a person in the system. Preprocessing and
feature extraction leads to storage in a data base that organizes the collection. To investigate the impact of forgery to the
authentication process, the inner and outer security is calculated using the handwriting data of genuine users and
specially trained forgers. During the authentication process the handwriting data, of users and forgers, is matched with
the enrollment data of the database after preprocessing and feature extraction. The matching uses the statistical method
of the biometric hash algorithm described in [8] and leads to a decision that defines the system’s outer and inner security
aspects.

data base
data acquisition
handwriting preprocessing feature extraction matching decision
authentication process
data acquisition
handwriting preprocessing feature extraction
metadata
enrollment process

Figure 1: Function of enrollment and authentication process

In this work we will introduce in the terms of metadata. The classes of technical and non-technical, the biological and
cultural, metadata are presented. Furthermore, the new class of conditional metadata is defined. Then, the handwriting
samples giving base for the analyses are defined and the different tasks of given, individual and creative handwriting
tasks are described. The experimental setup including error rates measurement and different forgery classes with focus
on the new metadata regarding forgery type blind meta attack are defined. Putting together metadata analysis and
handwriting data analysis of the intercultural working group CultureTech, three hypotheses suspecting intercultural
handwriting aspects are given. Evaluations of these hypotheses with respect to inner and outer security lead to our
distinct recommendations of best usable handwriting samples for differing user groups. Future work and possibly
applications conclude this paper.
SPIE-IS&T/ Vol. 6072 60720H-2